But which day? Such a second would probably work for everyday purposes (just like to most of us a second is 1/86400 of a day) but it would be woefully inadequate as a universal fundamental unit of time. And defining the second in terms of something like "the length of the solar day at Greenwich meridian on Jan 1 2020" prevents anyone from ever reproducing the exact value based on their own measurements, which is one of the big reasons for defining fundamental units in term of physical constants.
Yep, things leap seconds have been used for. The moon also constantly robs Earth angular momentum via tidal interactions.
But an even bigger issue is that the sidereal day (one 360° rotation of Earth as measured against distant stars) is not 24 hours but roughly four minutes less. And the length of the solar day also varies over the year due to the slight eccentricity of Earth’s orbit—days near perihelion are slightly longer than near aphelion. And then there are the higher-order effects caused by gravitational interaction with other planets...
One can arbitrarily define a second as anything and then say that one UTC day usually has 100000 seconds and some are longer or shorter. It is just a matter of scaling few unit definition constants by 10000/86400.
